GPT-4o’s Upgrade is Kinda Wild (Coding, Images, Unhinged Mode?)

GPT4 received an update with a new unhinged mode that can be activated by asking it to activate unhinge mode. This update improves coding instruction following and freedom, resulting in a significant performance boost on various leaderboards. The model now ranks as the second best overall, with less filtering when the unhinge mode is activated.

The update enhances the model's capabilities in coding tasks by reducing filtering and improving performance, as noted by Sam Alman. The model shows promise in handling coding prompts and provides satisfactory results compared to previous versions like Gemini 3.5.

Experimenting with a prompt involving creating detailed animations for TV channels using GPT4. The model successfully generates code for a working TV screen with channel titles that change dynamically, inspired by real TV channels. Gemini's code production and creativity are highlighted in this test.

Testing GPT4's ability to complete code prompts and facing challenges in getting complete code outputs, especially for complex tasks. Limitations in generating code for intricate prompts and the need for improvements in code completion functionality are discussed.

Exploring the model's ability to generate visual images, such as an SVG of a Pelican riding a skateboard. Despite some errors and limitations, the model shows potential in creating visual content based on input prompts and instructions.

Engaging in an ethics and decision-making prompt scenario, inspired by the classic trolley problem. Discussing the model's responses and insights on ethical dilemmas, showcasing its capabilities in tackling thought experiments and philosophical scenarios.

Reflecting on the model's performance, improvements, and implications of the recent updates. Observing the tone, style, and responses generated by GPT4.5, highlighting the unique features and developments in AI language models.